Update app.py

app.py

@@ -1,23 +1,26 @@
 from flask import Flask, render_template, request, jsonify
+import os
 import torch
+import speech_recognition as sr
 from transformers import pipeline
 from gtts import gTTS
-import os
 import re
-from waitress import serve  # Use Waitress for Production
+from pydub import AudioSegment
+from pydub.silence import detect_nonsilent
+from waitress import serve
 
 app = Flask(__name__)
 
-# Load Whisper Model for English Transcription
+# Load Whisper Model for Accurate Speech-to-Text
 device = "cuda" if torch.cuda.is_available() else "cpu"
-asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-base", device=0 if device == "cuda" else -1)
+asr_model = pipeline("automatic-speech-recognition", model="openai/whisper-small", device=0 if device == "cuda" else -1)
 
 # Function to generate audio prompts
 def generate_audio_prompt(text, filename):
     tts = gTTS(text=text, lang="en")
     tts.save(os.path.join("static", filename))
 
-# Generate audio prompts
+# Generate all required voice prompts
 prompts = {
     "welcome": "Welcome to Biryani Hub.",
     "ask_name": "Tell me your name.",
@@ -28,7 +31,7 @@ prompts = {
 for key, text in prompts.items():
     generate_audio_prompt(text, f"{key}.mp3")
 
-# Mapping for correctly converting spoken symbols to text
+# Symbol mapping for better recognition
 SYMBOL_MAPPING = {
     "at the rate": "@",
     "at": "@",
@@ -41,12 +44,24 @@ SYMBOL_MAPPING = {
     "space": " "
 }
 
-# Function to clean and process transcribed text
+# Function to clean and format transcribed text properly
 def clean_transcription(text):
     text = text.lower()
+    text = re.sub(r"\s+", " ", text).strip()  # Remove extra spaces
     for word, symbol in SYMBOL_MAPPING.items():
         text = text.replace(word, symbol)
-    return text
+    return text.capitalize()
+
+# Function to detect speech duration and avoid cutting words
+def trim_silence(audio_path):
+    audio = AudioSegment.from_wav(audio_path)
+    nonsilent_parts = detect_nonsilent(audio, min_silence_len=700, silence_thresh=audio.dBFS-16)
+    
+    if nonsilent_parts:
+        start_trim = nonsilent_parts[0][0]
+        end_trim = nonsilent_parts[-1][1]
+        trimmed_audio = audio[start_trim:end_trim]
+        trimmed_audio.export(audio_path, format="wav")  # Save trimmed audio
 
 @app.route("/")
 def index():
@@ -62,13 +77,16 @@ def transcribe():
     audio_file.save(audio_path)
 
     try:
-        # Transcribe audio to text with Whisper
+        trim_silence(audio_path)  # Trim silence before processing
+        
+        # Force Whisper to transcribe only in English
         result = asr_model(audio_path, generate_kwargs={"language": "en"})
         transcribed_text = clean_transcription(result["text"])
+        
         return jsonify({"text": transcribed_text})
     except Exception as e:
         return jsonify({"error": str(e)}), 500
 
-# Run Waitress Server for Production
+# Use Waitress for Production Server
 if __name__ == "__main__":
     serve(app, host="0.0.0.0", port=7860)